AI Makes Us WAY Faster but Fries Our Focus (Here's What I Do About It) — Transcript

Full transcript

1. 0:00It's incredible how much we're able to
2. 0:01accomplish with AI these days. With my
3. 0:03coding agents, their harnesses, and my
4. 0:05second brain, I have never been more
5. 0:07productive in my life. And the common
6. 0:09sentiment seems to be everyone's feeling
7. 0:11this way, but there is a big underlying
8. 0:13problem we're all dealing with. You're
9. 0:15probably feeling this pain. And so,
10. 0:17let's talk about this today. And don't
11. 0:19worry, I won't leave you hanging. We'll
12. 0:20get into solutions as well. Now, I've
13. 0:22seen a lot of people word this well
14. 0:23across X and LinkedIn and YouTube, but
15. 0:25this tweet definitely takes the cake. AI
16. 0:28was supposed to make us work less, but
17. 0:29I've never been this tired in my life.
18. 0:32And he talks about how he's spinning up
19. 0:33a lot of agents in parallel throughout
20. 0:35the day, and that forces him to jump
21. 0:37from conversation to conversation. And
22. 0:39here's the kicker. He's probably context
23. 0:41switching a hundred times more than he
24. 0:43did five years ago. And I would even say
25. 0:46two to three years ago. And so, by the
26. 0:48end of the day, my brain is fried,
27. 0:50completely toasted, nothing left, not
28. 0:52even for the people that I love. Now,
29. 0:54he's got to be exaggerating for effect,
30. 0:56at least a little bit, but this problem
31. 0:57is real. And maybe he's not even
32. 0:59exaggerating. He calls it the fried
33. 1:01brain effect. Thought it was just me,
34. 1:03but almost every founder I talked to
35. 1:04lately says the exact same thing. We got
36. 1:06superpowers, and somehow the pressure
37. 1:08only went up. And I am feeling this
38. 1:11pain. There's a good chance you are as
39. 1:12well. And to put it quite simply, the
40. 1:14problem here is the context switching,
41. 1:16because we are dedicating very small
42. 1:18amounts of attention at a time to each
43. 1:19problem, going between them constantly.
44. 1:22And it's killing our attention spans as
45. 1:24well. Kyle Cook on YouTube, he talked a
46. 1:26lot about this recently. He gave this
47. 1:28quote from Cal Newport. The ability to
48. 1:30focus is becoming the scarcest commodity
49. 1:32of the 21st century. And the best part
50. 1:34here is this quote is even before AI.
51. 1:37And so, AI is just making this problem
52. 1:39way worse. And what it comes down to is
53. 1:41our cognitive bandwidth doesn't
54. 1:43parallelize. Humans are not made to
55. 1:46context switch as much as AI is forcing
56. 1:48us to, or as much as we think AI is
57. 1:51forcing us to. We're going between these
58. 1:53tasks constantly. We're never dedicating
59. 1:56a solid chunk of time to a problem, and
60. 1:59so we're never getting in that flow
61. 2:00state where we're truly focused on
62. 2:02anything. Now, for me personally, and
63. 2:04I'm sure you can resonate with this as
64. 2:05well, even a couple of years ago, I
65. 2:07wasn't using coding agents. I didn't
66. 2:09have a second brain. I was still doing
67. 2:11deep focus work on tasks throughout my
68. 2:13day. And so, something that I genuinely
69. 2:15miss is turning the music on, jamming
70. 2:18out as I code for 4 hours straight, just
71. 2:20super focused in that flow state,
72. 2:22working on a problem. And I just simply
73. 2:24never do this anymore because at least
74. 2:26to me, it doesn't feel practical. I need
75. 2:27to make sure that I'm being as
76. 2:28productive as possible with what AI is
77. 2:31enabling me to do. And so, I'm working
78. 2:33on my second brain and my YouTube
79. 2:34content, maybe I'm recording a video for
80. 2:36a chunk of the day and then going back
81. 2:38to course planning, but I'm never doing
82. 2:39one thing for multiple hours because I
83. 2:41send a request to my coding agent, I
84. 2:43have a bit of a conversation with it,
85. 2:45and then it goes and does some work. And
86. 2:46so, then I switch over to the next agent
87. 2:48work on the next thing. And so, I
88. 2:49constantly have many agents running in
89. 2:51parallel because otherwise, I feel like
90. 2:53I'm not being as productive as I can be.
91. 2:55And this is exactly the problem that
92. 2:56Tibo's talking about in the tweet. The
93. 2:58pressure only went up because if you can
94. 3:00ship 10 times more, you feel like you
95. 3:02have to ship 10 times more every single
96. 3:05day. There's almost this guilt that
97. 3:06comes up where we're simply waiting for
98. 3:08an agent to finish because we know
99. 3:09there's something else we can start in
100. 3:11another conversation. The sponsor of
101. 3:13today's video is Revolut Business. My
102. 3:16second brain handles a good chunk of my
103. 3:18business now, but something that I've
104. 3:20delayed adding for a while is allowing
105. 3:22it to actually pay for some things. For
106. 3:24obvious reasons, I'm not comfortable
107. 3:26handing an AI agent my company card, at
108. 3:29least not without some very strict
109. 3:30controls. And no, having spending limits
110. 3:33in your prompts is not good enough. So
111. 3:35now, my second brain opens its own card
112. 3:37instead. Take a look at this. This is a
113. 3:40real API call to the Revolut Business
114. 3:42API, exactly what my second brain does
115. 3:44internally. It opens up a card, and what
116. 3:47it gets back is a virtual card with a
117. 3:49$50 weekly spend limit. It's locked to
118. 3:51purchasing software, locked to the US,
119. 3:54and the card expires at the end of the
120. 3:56month. And those limits, they're not in
121. 3:57the code, they're not in the prompts,
122. 3:59they're on the card itself. This is the
123. 4:01virtual card assigned to my second
124. 4:03brain. We're looking at it here in the
125. 4:04Revolut Business app. And so, if my
126. 4:06second brain uses this card to try to
127. 4:08buy something for $60, it can't happen.
128. 4:11There's no way it can work around the
129. 4:12guardrails, the card will decline,
130. 4:14guaranteed. And whenever my second brain
131. 4:16uses this card, Revolut Business fires a
132. 4:18webhook to my service. And so, it comes
133. 4:21into my system so I can handle it as I
134. 4:23need, setting up alerts, documenting it,
135. 4:26whatever it takes so I don't have to
136. 4:27monitor my second brain, or find out
137. 4:29about the charges later. And that's the
138. 4:31piece that I was missing. Now, I can
139. 4:32give my agent a budget without giving it
140. 4:34access to my entire account. Revolut
141. 4:37Business is giving you $200 when you
142. 4:39open up an account with my link, or with
143. 4:41the QR code on screen, and you add at
144. 4:43least a cent. I'll have a link to them
145. 4:45in the description. Now, Tibo and
146. 4:46everyone else, they've already covered
147. 4:48this problem. So, I want to share my
148. 4:49pers- pective with you on this, but I
149. 4:51also want to cover something that I
150. 4:52haven't really heard before, how we're
151. 4:54stuck in the in-between. And this is
152. 4:56also going to get into the solution that
153. 4:58will cover at the end of the video. And
154. 5:00so, it's really interesting because
155. 5:01before AI, it was very easy for us to
156. 5:04focus on just one thing for a very long
157. 5:06period of time, and still be considered
158. 5:08ultra productive. But now, ultra
159. 5:10productive means having as many agents
160. 5:11running as we possibly can, or at least
161. 5:13that's what it feels like. Because yes,
162. 5:15there are certain pieces of work we can
163. 5:16hand to agents to go through for the
164. 5:18entire day, but for the most part, LLMs
165. 5:21and harnesses aren't capable enough for
166. 5:22us to not babysit. We have all these
167. 5:24agent sessions running, but we're
168. 5:26constantly monitoring and providing
169. 5:28input. And so, that is what forces us to
170. 5:30switch our context so much throughout
171. 5:32our day. But, we're going to get to a
172. 5:35point. I mean, naturally, the
173. 5:36capabilities of LLMs are going to evolve
174. 5:38where we are, for most tasks, going to
175. 5:41be able to let the agent just rip
176. 5:42through it for the entire day. And what
177. 5:44that means is we have to babysit agents
178. 5:46less. If we can have an AI agent truly
179. 5:50act as an employee where it can take
180. 5:51autonomy for a full day for most things,
181. 5:54then this problem starts to go away and
182. 5:56we can become quite focused on one thing
183. 5:59throughout the day while we have our
184. 6:00background agents ripping through stuff.
185. 6:03And so, we're going to get there, but
186. 6:04we're definitely not there yet. We're
187. 6:06doing this babysitting and until we get
188. 6:08to this point, even as we make our
189. 6:10harnesses more and more capable, we just
190. 6:13feel the need to spin up more and more
191. 6:14agents. I mean, that temptation is
192. 6:16always going to be there. When you make
193. 6:17your second brain or your harness or
194. 6:19whatever just a bit more capable, your
195. 6:21agents are able to run a bit longer,
196. 6:23then instead of waiting, you just want
197. 6:24to add another session. But that does
198. 6:25lead to you context switching between
199. 6:27more conversations throughout the day.
200. 6:30And so then your day starts to look like
201. 6:31this. This is the blue line before we
202. 6:33had agents. We're focusing on one
203. 6:35problem or just a few problems
204. 6:37throughout the day, but now we get
205. 6:39started. We start a couple of
206. 6:40conversations, we have a couple of
207. 6:42things going in parallel, and then we
208. 6:43get that notification. All right, time
209. 6:45for us to provide some input or keep the
210. 6:47conversation going. So we go back and
211. 6:49that context switching that kills our
212. 6:51focus. And then we're waiting for the
213. 6:52agents a little bit, but we feel that
214. 6:54guilt come on. Like, okay, we can't just
215. 6:55be waiting here. I know there's more I
216. 6:56can do with another agent conversation.
217. 6:58And so we start that. And then
218. 7:00throughout the day, we're watching,
219. 7:01supervising, never focused on one thing,
220. 7:04and that's what leads us to being
221. 7:05exhausted at the end of the day. Now, I
222. 7:07want to be clear here. I wouldn't go
223. 7:09back to what I was doing two years ago.
224. 7:12As much as it sounds nice, enjoy the
225. 7:15productivity boost of AI. Like I'm able
226. 7:17to accomplish so much in my business
227. 7:19where I could never see myself going
228. 7:21back as much as this is a problem that I
229. 7:23want to work on a solution for. And I
230. 7:24hear this all the time. We want a
231. 7:26solution without going back to this
232. 7:28because AI is here to stay. That our
233. 7:30productivity gains, they make it worth
234. 7:32it. We're not going back. It is an
235. 7:33incredible thing, but this also kind of
236. 7:35sucks. Like I really don't like seeing
237. 7:37what this does to my focus. And I got to
238. 7:39be be sometimes my attention span. And
239. 7:42so, a lot of times I will procrastinate
240. 7:44context switching more. It's kind of
241. 7:46silly, but sometimes I will
242. 7:48procrastinate by playing chess cuz I
243. 7:50just have that craving to focus on
244. 7:52something for a little bit instead of
245. 7:54just dictating to my agents all day. Who
246. 7:56would have thought that chess would
247. 7:57become my ultimate procrastination tool,
248. 7:59but here I am. Now, I'm not doing it a
249. 8:01ton, but like yeah, a few times each day
250. 8:03I'm going to pop over to Lichess or
251. 8:04chess.com and just play a couple of
252. 8:06games. It's really strange cuz it's just
253. 8:08like how I want to go to the gym at the
254. 8:10end of the day because I've been
255. 8:11standing and sitting at my desk all day
256. 8:13and not moving very much. And so,
257. 8:14mapping that to this, I crave these
258. 8:17hobbies where I'm focusing deeply like
259. 8:19reading a book or playing chess because
260. 8:21I'm not focusing as much at work. Or at
261. 8:23least I haven't been, but I am working
262. 8:25on solutions to this problem. And so,
263. 8:28that's what I want to talk about here.
264. 8:29Getting really practical like how can we
265. 8:30invest our time so that this becomes
266. 8:33less of an issue because we're going to
267. 8:34get past this in-between point. Let's
268. 8:36talk about that now. Because obviously,
269. 8:38even though there are a lot of benefits
270. 8:40to using AI in this way, I don't want to
271. 8:42sit idly by as these downsides partially
272. 8:45shape how my role looks. And so, I'm not
273. 8:47going to accept it, and you don't have
274. 8:49to either. Let's talk about how we can
275. 8:50invest our time to ease the burden of
276. 8:53this problem over time. And so, we
277. 8:55already talked about how when we improve
278. 8:57our system, we make our coding agents
279. 8:59more capable, it tempts us to just add
280. 9:02another session. And so, we can keep
281. 9:04working that way. We can just keep
282. 9:05adding more and more things in parallel,
283. 9:07making it so that we're context
284. 9:08switching between more and more
285. 9:09conversations over time. But, there is
286. 9:12an alternative. Now, it takes a bit of
287. 9:14investment in not being quite as
288. 9:15productive now, but it's going to lead
289. 9:18to solving this problem eventually. Now,
290. 9:20really what this comes to is instead of
291. 9:22just opening up more sessions in
292. 9:23parallel, we take a step back to think
293. 9:25how we can make our harness more
294. 9:27autonomous. And so, you are sacrificing
295. 9:30short-term productivity by doing this
296. 9:32because more sessions means more things
297. 9:34directly outputting for your business.
298. 9:36Here, we're taking a step back and
299. 9:38spending some time on the larger
300. 9:39harness, but it's going to pay dividends
301. 9:41very, very quickly because we're
302. 9:43figuring out how can we make it so our
303. 9:45agent works more like we work, validates
304. 9:47like we validate, so we can trust it
305. 9:49more and more to work on longer running
306. 9:51tasks. And so we get to the point where
307. 9:53we truly have these background processes
308. 9:55that we can just let go for maybe even
309. 9:57the entire day. And so we're still going
310. 9:59to have decisions come back to us so we
311. 10:01can approve things, but it's going for
312. 10:03longer and we're batching the decisions
313. 10:05that we make and the babysitting that we
314. 10:07do so that we do have to contact switch
315. 10:09less. And we do that by working on our
316. 10:12larger harness. So I'm talking about our
317. 10:13rules and our skills, but even more
318. 10:16importantly that larger workflow of
319. 10:17having the agent plan, implement, and
320. 10:19validate and go through that loop by
321. 10:21itself on the larger tasks that we give
322. 10:23it. And so we define the intent up front
323. 10:26with the agent. We spend a lot of time
324. 10:27having it grill us and figuring out how
325. 10:29we can get on the same page for
326. 10:31everything and then we let it just rip
327. 10:33through everything. And that's exactly
328. 10:34what I'm building with my software
329. 10:36factory. And the reason I'm working on
330. 10:38the software factory is because I'm
331. 10:40trying as quickly as possible to get
332. 10:42past this sort of awkward in between
333. 10:44where I really do have my agents working
334. 10:46quite autonomously on a lot of the work
335. 10:48that I have for them. Now, when I talk
336. 10:50about these background processes, I am
337. 10:52not just talking about {slash} goal
338. 10:54built into something like Codex because
339. 10:56that is not good enough. We need more
340. 10:58customization because you need to be
341. 10:59thinking, "How do I have to babysit and
342. 11:02monitor my sessions?" The work is very
343. 11:04custom to you. And so how we codify this
344. 11:06sort of process into the larger harness
345. 11:09is also very custom to you. And so it's
346. 11:11not enough to just do {slash} goal and
347. 11:13say that you've solved this entirely
348. 11:15because you're going to have to babysit
349. 11:16a lot at the end when it screws up
350. 11:18because it doesn't know how you work and
351. 11:20how you validate, right? Like that's why
352. 11:22I'm building the software factory for
353. 11:23myself with a lot of things that I'm
354. 11:25working on now. Now, also you might not
355. 11:27be interested in building the software
356. 11:29factory yourself at all. And I totally
357. 11:31get that. I understand that's not for
358. 11:33everyone. But if you are more of a
359. 11:34builder at heart and you want the best
360. 11:37results possible as soon as possible, my
361. 11:40call to action here is to build the
362. 11:41software factory and understand how this
363. 11:43kind of system works. And I cover this
364. 11:45in a full video on my channel that I'll
365. 11:47link to right here where I go into like
366. 11:49how I plan the larger release with my
367. 11:51agent and then I have it autonomously
368. 11:52split into issues and go through the
369. 11:54loop with a whole manager agent
370. 11:55overseeing everything. Like this is
371. 11:57going to become the standard in the next
372. 11:59year or two for how everyone is building
373. 12:01software because we do all the planning
374. 12:03up front and then we let it go for an
375. 12:05entire day or even multiple days and we
376. 12:07trust the process because we've
377. 12:09engineered it custom to how we work
378. 12:11specifically. And so what we get at the
379. 12:13end is a bunch of work that we can
380. 12:15review in batch and test in some kind of
381. 12:17staging environment. It's a lot less
382. 12:19context switching because everything in
383. 12:21the middle is running autonomously. And
384. 12:23even if you don't want to build the
385. 12:24factory yourself, you're going to get to
386. 12:26the point where there are solutions
387. 12:27available to you to run it out of the
388. 12:29box that make it very easy to customize
389. 12:32to your work. And so if you build the
390. 12:34factory yourself like I'm doing, we're
391. 12:35going to get there faster, so there's
392. 12:37still very much a benefit here and it'll
393. 12:38be even more custom. But if you're less
394. 12:41of a builder or you're more of a vibe
395. 12:43coder and you just want to be using
396. 12:44these coding agents as they are out of
397. 12:46the box, you'll still get there
398. 12:47eventually. And so if that's you, if
399. 12:49anything this video is just helping you
400. 12:50understand like the awkward in between
401. 12:52we're in now and where we're going to
402. 12:54go. And I still think that's quite
403. 12:56valuable even if you're not going to
404. 12:57action on this right away. Because we
405. 12:59all are going to get to the point where
406. 13:01we have this sort of primary session
407. 13:03where we're doing the high-level
408. 13:04planning with our agents and then we
409. 13:05have something like a software factory
410. 13:07doing a lot of the heavy lifting behind
411. 13:09the scenes where we really only have to
412. 13:11check in maybe like once a day and that
413. 13:12is going to solve the problem of context
414. 13:15switching at least as much as we
415. 13:16possibly can. And really the biggest
416. 13:18thing that's going to solve our days
417. 13:19looking like this, constantly baby
418. 13:21sitting our agents, is getting them to
419. 13:23the point where they really act as
420. 13:25trained employees. The harness
421. 13:27engineering is the training. We're
422. 13:28codifying how we work so the agent can
423. 13:30just run for a day or multiple days. And
424. 13:33with tools like Muse and Grokbot, we're
425. 13:35starting to see that kind of autonomy,
426. 13:37but it's still pretty limited for the
427. 13:39things it can truly accomplish reliably.
428. 13:41But we're already starting to see that.
429. 13:43Right? Like the in-between, there's
430. 13:45already light at the end of the tunnel
431. 13:46here. We're never going to go back to
432. 13:48fully focusing for hours and hours just
433. 13:50jamming out to music as we're coding,
434. 13:52but we are going to get to this sort of
435. 13:53like we focus on one primary session at
436. 13:55a time. And another thing that's going
437. 13:57to be really helpful for this is getting
438. 13:59LLMs that have near instant responses.
439. 14:02Because we're also starting to see this
440. 14:04with LLMs and hardware that give us
441. 14:06hundreds of tokens per second. And so
442. 14:08when we have near instant responses,
443. 14:11that's going to allow us to
444. 14:12realistically live for the most part in
445. 14:14this primary session where we're doing
446. 14:16all of the high-leverage tasks for our
447. 14:17business like the planning and
448. 14:19validation around the things that we're
449. 14:20building. And so going back to this part
450. 14:22of the diagram here, we're not going to
451. 14:23be tempted to kill our attention by
452. 14:25switching to another conversation
453. 14:27because that primary one we're going to
454. 14:28be getting the responses fast enough
455. 14:30where we can really just live here and
456. 14:31then constantly be delegating things to
457. 14:34these larger workflows that we just let
458. 14:35run in the background. And so this kind
459. 14:37of becomes our orchestrator where we're
460. 14:38doing our planning, checking on the
461. 14:40larger work once a day or whatever when
462. 14:42we need to, but we don't have to be
463. 14:44going between all these different agent
464. 14:46sessions. Now a lot of this is
465. 14:48predictions for the future. And so who
466. 14:50knows what's really going to happen in
467. 14:51the next year or two, even the next 6
468. 14:53months. But I'm just trying to share the
469. 14:55awkward in-between we're in now, what
470. 14:57I'm building towards with the software
471. 14:59factory to solve it, and then also
472. 15:00things that I'm hoping for like near
473. 15:02instant responses to make it so we can
474. 15:04really live in one conversation and
475. 15:06conversate one-on-one with an agent for
476. 15:09a long period of time. And I guess
477. 15:11that's going to become the sort of new
478. 15:12focused work. And so I will forever miss
479. 15:15jamming out to music for 4 hours as I
480. 15:17code, but at least we're not going to
481. 15:19live in this in-between forever. Now one
482. 15:21last thing I want to mention here. I
483. 15:22brought up this idea to my Dynamis
484. 15:24community and a couple of really smart
485. 15:26guys there said, "Cole, this makes
486. 15:27sense, but what if we just start context
487. 15:29switching between these different
488. 15:31factories?" So, we just kind of go to
489. 15:32that next level of productivity and
490. 15:34parallel work. And I got to say, that
491. 15:37might happen to some people. Like, I
492. 15:38don't have the answer to everything.
493. 15:39Maybe we really are on this hamster
494. 15:41wheel of AI forever. But, my main
495. 15:43argument here is hopefully we'll get to
496. 15:45the point where no matter what we want
497. 15:47to accomplish in our life or in our
498. 15:49business, we're able to delegate
499. 15:51everything through that primary session.
500. 15:53That is what I'm working towards myself.
501. 15:55And so, hopefully you found this
502. 15:56enlightening. Just understanding where
503. 15:57we're at right now with AI killing our
504. 15:59focus and the light at the end of the
505. 16:01tunnel. We're not going to be here
506. 16:02forever and these solutions are
507. 16:04something I'm working on actively. And
508. 16:06so, if you found this interesting, it
509. 16:08resonated with you, you're curious about
510. 16:10the software factory that I'm building,
511. 16:12come follow along. I'm building all this
512. 16:14in public and I'd really appreciate a
513. 16:15like and a subscribe. And with that, I
514. 16:18will see you in the next video.